IWE MCP
IWE keeps project memory in linked Markdown notes: a CLI, an editor server and the iwec MCP server let an agent read and edit them alongside you
- 1,806 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- The MCP server creates, updates, deletes and renames documents in the workspace, and edits go straight to files
- The `iwe_query` tool runs bulk `update` and `delete`, and `iwe_normalize` rewrites the formatting of every document in the graph
- The config created by `iwe init` holds a command that launches an external agent, `claude -p`
Install
- Works in
Manual install
brew install iwe-org/iwe/iweThen run `iwe init` in the notes folder. Other options: npm and cargo.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The MCP server creates, updates, deletes and renames documents in the workspace, and edits go straight to files
- The `iwe_query` tool runs bulk `update` and `delete`, and `iwe_normalize` rewrites the formatting of every document in the graph
- The config created by `iwe init` holds a command that launches an external agent, `claude -p`
This is third-party code. Review the repository files before installing.
What it does
IWE turns a folder of Markdown files into a graph of linked documents: specs, plans, decisions. One note can belong to several topics, and links between notes give backlinks. The folder becomes a workspace after `iwe init`, which creates a `.iwe` directory with a `config.toml` file. Four tools then work on the same files: the `iwe` CLI, the `iwes` language server for Neovim, Helix, VS Code and Zed, the `iwec` MCP server and the iWe app for Mac. There is no model in either the CLI or the app. The CLI searches notes (`iwe find`, with `--lexical` for a text search), shows the tree, retrieves documents with a chosen depth and backlinks (`retrieve`), creates and renames them with link updates, moves a section into its own document (`extract`), puts a link back inline (`inline`), counts statistics and brings formatting to one form (`normalize`). The `iwec` MCP server gives an agent 14 tools: search and retrieval, tree, statistics, creating, updating and deleting a document, renaming, extracting and inlining a section, normalization, attaching, and a free-form query in the IWE query language. It also serves resources such as `iwe://documents/{key}` and the prompts `explore`, `review` and `refactor`. The server reads `.iwe/config.toml` from the workspace root. For Claude Code there is a plugin: once installed, the `/iwe:init` command connects memory to a project.
Who it is for. For developers and writers who keep specs, plans and decisions in Markdown and want Claude Code, Codex or Cursor to read and extend those notes instead of retelling context every session.
Good fit when
- Every agent session begins with retelling the project, while the decisions and their reasons sit in notes the agent cannot reach
- You want memory you can open, correct and delete by hand, unlike the agent's own hidden storage
- You write notes in Neovim, Helix, VS Code or Zed and need link completion, backlinks and safe renaming
- A large document has to be split: `iwe extract` moves a section into its own file and leaves a link behind
- One set of notes should be read by different agents: Claude Code, Claude Desktop, Codex, Gemini and Cursor through MCP
Not a fit when
- You need vector and hybrid search by meaning: IWE describes its search as fuzzy, structural and link-based, while Basic Memory offers vector search
- You have no Homebrew, npm or Rust: installation goes through one of these three
- Your notes live in a cloud service rather than in a folder on disk: IWE works only with the files of a local workspace
- You want a ready cloud service with sign-in and sync across devices: IWE has no account, and you set up sync (git or another) yourself
Example request
Find the authentication decisions in my IWE notes and add to the spec why we dropped cookie sessionsMCP
- Transport
- stdio
- Authentication
- not required
Environment variables
IWE_DEBUG- A value of 1 turns on the server's debug logging to stderr
Limitations
- The latest version, 0.26.1, came out on 8 October 2026 and is numbered below 1.0. The project already has config format migrations with tests.
- The iWe app for Mac is downloaded from iwe.md and its code is not in the repository, which holds the diwe, iwe, iwec and iwes crates. Check the app's terms on the site.
- The `.iwe/config.toml` that `iwe init` creates names the command `claude -p` with a 120-second timeout by default. That launches an external Claude Code, since IWE has no AI of its own. Connecting an agent needs that agent's own setup and account.
- The `.iwe` directory marks the project root: without `iwe init`, an ordinary folder of Markdown files is not an IWE workspace.
- Apache-2.0 license.
How to disable. Remove the `iwe` block from your agent's MCP settings and uninstall the Claude Code plugin through `/plugin`. The notes stay ordinary Markdown files, and the `.iwe` directory with the config can be deleted. The CLI is removed with `brew uninstall iwe` or `npm uninstall -g @iwe-org/iwe`.
FAQ
Do I need an IWE account or my own model?
No. Neither the CLI nor the Mac app includes a model, and storing notes needs no IWE account. Notes live in a folder on your computer and can be kept in git. The agent needs its own setup and account.
Which MCP tools can change or delete notes?
Of the 14 `iwec` tools, `iwe_create`, `iwe_update`, `iwe_delete`, `iwe_rename`, `iwe_extract`, `iwe_inline`, `iwe_normalize` and `iwe_attach` write. The `update` and `delete` operations are also reachable through `iwe_query`. Only `iwe_find`, `iwe_retrieve`, `iwe_tree`, `iwe_stats` and `iwe_squash` read.
How do I connect memory to Claude Code?
Run `/plugin marketplace add iwe-org/skills` and `/plugin install iwe@iwe-org`, then run `/iwe:init` in the project. For Codex and other agents, add the `iwec` MCP server by hand as the guide on iwe.md describes.
Which editors does IWE support?
Neovim, Helix, VS Code and Zed through the `iwes` language server. It adds link completion, backlinks, search and safe renaming of notes. You can use IWE without an agent.
How do I run the MCP server without installing it?
Through npx: in the agent settings use `command: npx` and `args: ["-y", "@iwe-org/mcp"]`, and set `cwd` to the workspace path. An installed binary starts as `iwec` with no arguments.
Discussion
No questions or reviews yet.
Related
MCP serversWEEEK MCP with confirm-gated writes
weeek-mcp
A WEEEK MCP server that is read-only by default: tasks, attachments and comments via a browser session, writes are opt-in
WEEEK MCP with a read/write split
weeek-mcp-server
A 12-tool WEEEK MCP server split into 7 read and 5 write tools, so an MCP client can auto-approve the safe ones
WEEEK MCP with the knowledge base
weeek-mcp
A WEEEK MCP server: tasks via the public API and full knowledge-base CRUD via the internal API with a saved browser session
WEEEK plugin with 12 curated tools
weeek-mcp
A Claude Code and Codex plugin for WEEEK: 12 curated, not auto-generated tools, with the token in the OS keychain
A build script and templates that turn an Obsidian vault into an AI's working memory: markdown notes, no vector database
Turns session history of Claude Code, Codex, Cursor and Gemini CLI into a searchable wiki on Karpathy's LLM Wiki pattern, locally
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| IWE (this entry) | no | not required | 1,806 | |
| Basic Memory | no | not required | 4,403 | memory in local Markdown files with a graph, semantic search and an SQLite index |
| MCP Memory Service | no | not required | 2,054 | self-hosted memory for agents with semantic search and auto-consolidation |
README in short
The repository's front page explains the idea of shared memory for a person and agents and lists four ways to work: the Mac app, an editor, the CLI and Claude or Codex. Next come a Homebrew install, first commands, a Claude Code block with the plugin and a text you can paste into any agent for self-setup. The repository holds help for every CLI command, guides to configuration and the query language, schemas, speed benchmarks, crate tests and links to ready workspaces for development and marketing.